Immunotherapy for acute myeloid leukemia: from allogeneic stem cell transplant to novel therapeutics

Abstract

Immunotherapy in the form of allogeneic stem cell transplantation (SCT) plays an instrumental role in the treatment of acute myeloid leukemia (AML), with non-transplant modalities of immunotherapy including checkpoint blockade now being actively explored. Here, we provide an overview of the graft versus leukemia (GVL) effect in AML as a window into understanding the prospects of AML immunotherapy. We explore the roles of various cell types in orchestrating anti-leukemic immunity, as well as those contributing to the unique immune suppressive state of myeloid diseases. We discuss specific approaches to engage the immune system, while noting the challenges of the AML antigen landscape and the barriers to immune modulation. We review the potential for immunomodulatory agents in combination with cellular therapies, donor lymphocyte infusion, and following SCT. Finally, to address the challenge of minimal residual disease (MRD) following chemotherapy, we propose combination epigenetic and immunotherapy for the eradication of MRD.

Introduction

Patients with high-risk AML have a poor prognosis. High-risk AML can be defined by both biologic and clinical variables, including age, cytogenetics, mutational profile, exposure to prior genotoxic chemotherapy, and response to chemotherapy[1–8]. Once any AML patient has failed first-line approaches, there remains no standard of care for patients lacking targetable mutations[9]. Even with these failures, until recently, our standard approach to upfront therapy for AML induction and consolidation had remained static for decades.

For high-risk AML patients, allogeneic SCT (alloSCT) offers the best opportunity at cure[1]. AlloSCT is the most fundamental form of immunotherapy, in which the donor immune system recognizes and eradicates residual AML in the graft-versus-leukemia (GVL) effect[10]. However, there are significant limitations for alloSCT in high-risk AML. AlloSCT has the best outcomes in first complete remission (CR1), and many patients with high-risk AML do not achieve a first remission[1]. AlloSCT carries significant risks of treatment-related morbidity and mortality. Although reduced intensity conditioning regimens have expanded the number of patients eligible for allogeneic SCT, patients with advanced age are still more likely than younger patients to have co-morbidities precluding them from transplant. Even for those AML patients who are candidates, relapse after allogeneic SCT remains a significant challenge [1,11]. Recent data have also demonstrated that the presence of any detectable AML prior to transplant, or minimal residual disease (MRD), portends a poor prognosis and a high rate of relapse following alloSCT. These MRD positive patients may have an equally poor outcome as those who enter transplant with active AML[12], and represent an important area of unmet need for novel therapeutics. Thus, there is ample opportunity to improve outcomes in this difficult patient population.

Allogeneic Stem Cell Transplantation and Mechanisms of the Immune Responses to AML

The immunotherapeutic benefit of alloSCT was recognized with the observation that patients with higher rates of GVHD had lower rates of relapse[13]. As it was known that allogeneic T cells were major drivers of GVHD, it was hypothesized that these cells could also drive the GVL effect [14]. T cells recognize tumor cells through binding of the T cell receptor (TCR) to cognate major histocompatibility complex proteins (in humans, human leukocyte antigen, HLA) which present tumor antigen on the cell surface (Figure 1). This binding, along with positive costimulatory signals, activates the T cell to kill a tumor cell. This interaction can be exploited in several ways. First, mismatch at both minor and major histocompatibility antigens can drive a beneficial antitumor response in which donor T cells recognize a tumor cell as “foreign” in the recipient [13,15]. In HLA-matched sibling alloSCT, this is accomplished by disparate minor polymorphic non-self peptides, or minor histocompatibility antigens. By looking at large data sets, investigators have gradually improved donor-recipient HLA-typing to avoid GVHD yet preserve GVL [16]. To date, the majority of our understanding of the cellular response against AML comes from the decades of experience and improvements in allogeneic stem cell transplantation.

Figure 1. Overview of cellular mediators against AML blasts and potential pathways to engage them in vivo.

The various effector cells including NK cells, macrophages, effector and regulatory T cells are shown. Antibodies targeting tumor associated antigens (TAAs) on AML blasts can be in their native form, as antibody drug conjugates (ADCs), or generated as bispecifics to engage effector T cells. Dendritic cells, one of the most potent antigen presenting cells, can process antigens endogenously or provided through vaccination strategies in order to expand tumor-specific T cells.

Mobilizing the cellular response against AML

Multiple investigators have attempted to more specifically target T cells against commonly overexpressed tumor antigens in AML. Initial attempts at enhancing the antigen-specific T cell response to AML were based on vaccines eliciting tumor-specific T cells [18,19]. The basis for this approach was to provide defined tumor associated antigens to professional antigen presenting cells (APCs), such as dendritic cells, helping promote antitumor T cells (Figure 1). These included vaccines directed against minor histocompatibility antigens (HA-1, HA-2)[20], proteinase 3 (PR3) [21], Wilms tumor protein (WT-1)[19,22], or tumor cells overexpressing GM-CSF[23]. Given the largely immunosuppressive environment found in many malignancies including AML, not all vaccine responses have been sufficient enough to lead to in vivo expansion of antigen-specific T cells and tumor regression [24,25]. AML presents unique challenges for immunotherapy in that it is disseminated throughout the body at diagnosis and lacks the tumor draining lymph nodes key for orchestrating an effective immune response. The systemic nature of AML appears to induce a unique state of immunologic tolerance, an effect that is not seen when the same tumor is given subcutaneously[26]. Part of the immunosuppressive environment of AML can also be explained by defects seen in dendritic cells, which are important in stimulating the adaptive immune response. Kline and colleagues have demonstrated a role for CD8⍺ plasmacytoid DCs in mouse models of AML which are capable of cross-presenting leukemia associated antigens to T cells [27,28]. Patients with AML show both quantitative and functional defects in circulating blood dendritic cells which could hamper optimal cross-presentation [29]. More recently, our group has characterized a depletion of plasmacytoid dendritic cells with increasing levels of disease, indicating that a similar mechanism may be operative in humans[30].

As opposed to what was initially hypothesized, AML patients appear to have adequate numbers of T cells at diagnosis[31]. These T cells, however, are phenotypically different in that they express significantly higher levels of activation markers such as CD25 and CD69. Similar to what is found in pre-clinical models these T cells could appear activated in the setting of recognizing AML antigens but undergo abortive tolerance in the absence of DC stimulation[32]. In humans, this is supported by recent data demonstrating markers of T cell exhaustion (e.g. PD-1, TIM3, LAG3) and senescence at diagnosis, leading to divergent outcomes post-induction between responders and non-responders. Deep molecular profiling of responders has demonstrated upregulation of costimulatory markers and T cell signaling pathways indicating restoration of the T cell response[33]. Others have also shown increased levels of the inhibitory receptor PD-1 on T cells at the time of AML relapse, with a shift from naïve to effector memory populations, both supporting antigen-stimulated cells. These studies, however, saw no diminished function compared to healthy controls when these cells received ex vivo stimulation with mitogens [34]. In addition, more recent work found an increased frequency of Tregs in bone marrow aspirates from patients with AML, particularly in patients with greater than one relapse[31]. Given increased frequency of Tregs and increased expression of both exhaustion/inhibitory markers at diagnosis and relapse, there has been significant interest in the possible use of immune checkpoint blockade alone or in combination following induction chemotherapy.

Multiple immune checkpoint receptors including PD-1, OX40, and ICOS are expressed on T-cells in bone marrow aspirates from patients with newly diagnosed and relapsed AML[31]. Co-expression of specific immune checkpoint proteins (TIM-3 and LAG-3) with PD-1 is associated with T-cell exhaustion, and TIM-3 and LAG-3 are co-expressed more frequently with PD-1 on T-cells isolated from AML than T-cells from healthy donors[31]. PD-L1 is also expressed on some AML cells, although levels of PD-L1 expression have been variable across studies[35–39]. The largest case series shows that PD-L1 expression is low or undetectable on most cases of primary AML and expressed at low levels in only 16.3% (20/123) of cases at diagnosis[39]. However, smaller series suggest that PD-L1 expression may increase on AML blast cells during treatment and following relapse[35,36,40], and PD-L1 expression on AML cells is negatively correlated with outcome[36–38]. Pre-clinical mouse models also point towards the PD-1/PD-L1 pathway in immune evasion by AML[42–45], leading to several trials of checkpoint blockade in AML and MDS discussed in more detail below [46,47].

If tissue resident T cells are present but in a non-functional state, an alternative approach could be to isolate tumor specific T cells from patients, expand them to large numbers ex vivo and then re-infuse them into the patient. First developed at the NCI, adoptive T cell therapy has been used in many different clinical settings [48]. T cells directed against PR3, WT-1, mutated nucleophosmin, and TERT have been examined in AML[21,49,50]. Alternatively, T cells can be genetically modified with tumor-specific TCRs and expanded ex vivo. However, given that tumor-specific TCRs depend on precise TCR-HLA interactions, they are not broadly applicable as even the most common HLA-directed therapies against HLA-A2*01 are only found in ~20-30% of Caucasians. Therefore, alternative methods of targeting T cells to AML cells are needed. Given the success of chimeric antigen receptor (CAR)-expressing T cells in other hematologic malignancies, several groups have worked to identify appropriate target antigens in AML. Recently, Perna and colleagues took a non-biased proteomics approach and identified several highly expressed surface antigens in AML that may serve as promising targets for CAR-T cells in AML[51]. This would be an advance over current strategies targeting more broadly expressed antigens such as CD33 or CD123[52]. While these studies have had some success in patients with AML, much work needs to be done to optimize this promising therapy for this disease.

Natural killer (NK) cells, which are part of the innate immune system, also play an important role in the immune response against AML[53]. Following cytotoxic induction or maintenance therapy, NK cells are impaired and less cytotoxic towards AML blasts[54]. NK cells sense tumor cells differently than T cells in that they are not “restricted” to recipient HLA. This “missing-self” hypothesis was first demonstrated in mice where NK cells were able to lyse leukemia cells lacking HLA molecules[55]. In fact, lack of this recognition drives a strong anti-tumor response and this mismatch can be optimized in the SCT setting. NK cell activity in AML is driven primarily by the balance of activating and inhibitory signals delivered to the NK cell after it encounters the tumor. These signals are primarily driven through the family of killer immunoglobulin receptors, or KIR[56]. In AML, the importance of this interaction was first recognized clinically by Ruggeri and colleagues who found that in AML patients undergoing SCT, donors expressing KIR receptors not recognizing HLA ligands in the host had improved disease free survival and lower rates of GVHD[57]. Given the role of donor NK cells in alloSCT, Miller and colleagues pioneered the use of allogeneic NK cell infusions into patients with hematologic malignancies. They found that adoptive transfer of haploidentical NK cells with IL-2 infusion led to complete remissions in 5 of 19 patients with poor-risk AML. The patients who responded had an alloreactive NK cell repertoire in the graft vs host direction[58]. Based off these data, a number of adoptive cellular approaches have aimed to optimize NK cells for the treatment of AML[59,60]. While there are several pathways to manipulate in the cellular response, it remains important to consider the negative regulators that naturally limit these processes as another point of therapeutic intervention.

Negative regulators of the immune response to AML

In AML, both myeloid derived suppressor cells (MDSCs) and regulatory T cells play pivotal roles[64]. MDSCs are derived from hematopoietic progenitor cells otherwise capable of giving rise to monocytes, macrophages, neutrophils or granulocytes. MDSCs can be induced by the cytokines GM-CSF, IL-1, IL-6, and IL-10, all of which readily exist in the tumor microenvironment[65]. There are currently no good strategies for targeting MDSCs within the tumor microenvironment, partly owing to our lack of understanding of this ill-defined population of cells. MDSCs have been shown to create excess reactive oxygen species (ROS) within the tumor microenvironment inhibiting tumor reactive CD8 T cells[66].

T regulatory cells (Tregs) are important cells in maintaining tolerance to self, and thus play a major role in cancer development and immune suppression. As mentioned above, Tregs are increased in patients with AML, especially in the relapsed setting[31]. Tregs come in two forms, natural or induced. Natural Tregs develop in the thymus in response to high affinity interactions with medullary thymic epithelial cells expressing self-antigens and account for 5-10% of peripheral blood CD4 T cells. Peripheral Tregs can be induced from CD4 T cells in response to cytokines, such as TGF-β or IL-2 induced during inflammation and express high levels of the checkpoint CTLA-4. Tregs help control autoreactive T cells and thereby curtail autoimmune responses. This inhibitory mechanism can be usurped in cancer to help promote tolerance to certain tumors, including AML[31]. Studies in mice have shown that AML-associated Tregs can also block CTL responses by modulation of the PD-1/PD-L1 axis and that Treg depletion can lead to eradication of AML following PD1 blockade[67–69]. Soiffer and colleagues conducted a phase I trial looking at the use of CTLA-4 based checkpoint blockade in alloSCT recipients after relapse with Hodgkins lymphoma, multiple myeloma, or AML (including leukemia cutis and myeloid sarcoma)[70]. In twenty-one patients treated at 10 mg/kg ipilimumab, the overall response rate was 33%. 42% (5/12) of patients with AML achieved a complete response, with some of the responses durable at 8 month follow-up. Examination of immune correlates revealed that peripheral blood T cells peaked at 4 weeks after initiation of ipilimumab. Interestingly, peripheral Tregs decreased, with the ratio of Treg/Teff cells decreased in up to 41% of the 15 patients analyzed. This is consistent with the previously described mechanism of CTLA-4 blockade, in that it not only “takes off the brake” of T cells but also depletes CTLA-4 expressing Tregs in an Fc-dependent manner[71]. As one would expect, CTLA-4 blockade following alloSCT does have toxicity. Post-transplant ipilimumab was associated with immune-related adverse events (irAE) typical for ipilimumab in 6/28 patients (21%), including pneumonitis, diarrhea/colitis, and ITP. These irAEs were generally reversible with steroids, although one patient died of grade 3 colitis and grade 4 pneumonitis. Post-transplant ipilimumab also led to worsening of acute and chronic GVHD. Another study found similar changes in peripheral T cell counts (increased T cells, decreased frequency of Tregs) following CTLA-4 blockade as well as irAEs without evidence of GVHD, however, at lower doses of 3 mg/kg[73]. These studies demonstrate the need for a better understanding of immune reconstitution following alloSCT to more specifically modulate anti-leukemia efficacy.

Vaccine therapy revisited

While vaccination can potentially induce both T- and B-cell mediated immunity, prior attempts to vaccinate against single AML antigens did not lead to durable remissions. Thus, approaches providing a polyclonal response targeting multiple patient-derived antigens were hypothesized to be more effective against resistance and/or relapse. More recently, new approaches have re-invigorated this field with impressive responses. Avigan and Rosenblatt evaluated the use of an autologous tumor vaccine in 17 patients with high risk AML following remission induction and consolidation chemotherapy[74]. The vaccine consists of pre-treatment AML blasts from patients fused with patient derived DCs creating a hybridoma. Following vaccination, antigen-specific CD4 and CD8 T cell responses were induced against tumor antigens MUC1, WT1, and NY-ESO1. Compared to traditional responses in chemotherapy only treated patients (15-20% 2 year survival), twelve of the 17 (71%) patients remain in remission 5 years following initial treatment. There are now ongoing multi-center trials testing this approach. Other groups have attempted alternative methods to individualize recombinant vaccines on a per-patient basis. Using massively parallel sequencing to identify unique antigens in melanoma patients, investigators generated peptide vaccines with up to 20 personalized neoantigens admixed with toll-like receptor (TLR) agonists[75]. While this proof-of-concept study was a major advance in the field of vaccinology, it has yet to be extrapolated to other “cold” tumors where neo-antigen burden is substantially lower, including AML. The labor-intensive nature of such individualized cellular or recombinant products also presents significant challenges in scaling this approach for multiple patients.

The humoral response to AML

While the role of cellular immunity in controlling AML has been well described, much less is known about the antibody response. Some early anecdotal studies reported AML remissions in patients with hypergammaglobulinemia in the setting of transfusion or infections[76]. Others have worked to profile serum of AML patients for tumor associated antigens[77]. More recently, Hazenberg and colleagues took a novel approach to identify antibodies in patients with AML. Although the graft-vs-leukemia effect was thought to be primarily driven by T cells, they asked the simple question of whether or not protective antibodies can also be identified in patients following a successful allogeneic stem cell transplant. To do this this, they screened serum from transplant patients with high-risk disease at diagnosis who had survived past 5 years. By screening patient sera against AML lysates and following identification of positive clones, antibodies were isolated from these B cell clones[78]. Interestingly, they identified antibodies to a shared spliceosome protein U5 snRNP200 which is aberrantly expressed on the cell surface in AML patients. Furthermore, these antibodies showed activity against AML cells both in vitro and in vivo. They have gone on to characterize many other targets they have identified which could provide a new repertoire of antibody-based therapeutics for AML[79]. These data provide strong evidence that in combination with enhanced T cell activity, protective antibodies could provide durable control against high risk AML.

Although identification of these AML-specific antibodies provide important new targets for novel therapies, investigators have been using recombinant antibody technology to target AML since the early 2000s. These include both recombinant IgGs alone, as antibody drug conjugates (ADCs), or more recently bi-specific T cell engagers (BITEs), each of which have been extensively reviewed elsewhere[80].

Checkpoint blockade in the treatment of AML

The use of immunotherapy has changed the way we approach many cancers. Many immunotherapeutic drugs target inhibitory receptors and signaling pathways such as PD-1/PD-L1 and CTLA-4 that regulate immune cell activation in response to antigen stimulation, an approach described as immune checkpoint blockade[81–84]. Although the efficacy of immune therapy in hematologic malignancy has been established for decades in the form of allogeneic stem cell transplantation, the development of immune checkpoint inhibitors in hematologic disease has lagged behind solid tumors[46,47,85]. One notable exception involves the remarkable activity of single-agent PD-1 blockade in multiply relapsed/refractory Hodgkin lymphoma, where nivolumab yielded an overall response rate of 87% and CR rate of 17%[46,86]. In diseases such as leukemia, early-phase results from checkpoint inhibitor monotherapy have been less robust but this continues to be an evolving landscape[87]. One approach to improve response rates has been to combine multiple immune checkpoint inhibitors, though at the cost of increased toxicity from immune-related adverse events (irAEs)[82]. Another potential barrier to efficacy of checkpoint blockade is that de novo AML carries few mutations[88,89]. Mutational burden likely plays an important role in response to checkpoint blockade in promoting a specific neo-antigen landscape[90–92], thus ways to improve upon the recognition of AML blasts by a patient’s immune system are needed.

Epigenetic Therapy as Immmunotherapy

Epigenetic modifying drugs such as azacitidine and decitabine are commonly used in the treatment of patients with AML. They inhibit DNA methyltransferases and serve as hypomethylating agents (HMAs), altering patterns of gene expression. Although their mechanisms of action are not fully understood, HMAs provide some survival benefits to specific populations of AML and high-risk MDS patients, particularly elderly patients who are not candidates for intensive regimens. Pre-clinical data suggests that combining HMAs with immune checkpoint inhibitors in leukemia and other malignancies may enable synergistic activity and facilitate immunotherapy (Figure 2)[93–98]. HMAs and histone deacetylase (HDAC) inhibitors upregulate a variety of genes, including previously silenced leukemia antigens that could serve as targets for AML-specific immune responses. For example, treatment of AML cells with azacitidine, the HDAC inhibitor valproic acid, or the novel HMA SGI-110 induces expression of immunogenic cancer/testis antigens (CTAs) such as NY-ESO-1 and MAGE-A, and MAGE-specific cytotoxic T-lymphocyte (CTL) activity has been associated with major clinical responses[99–102]. Inhibitory immune checkpoint proteins including PD-L1, PD-L2, PD-1 and CTLA-4 can also be upregulated by treatment with HMAs[103]. Treatment with HMAs may therefore augment both tumor-specific immune responses by upregulating neo-antigens, as well as impair tumor-specific immune responses by upregulation of immune checkpoint proteins. However, such epigenetic drug-mediated immune checkpoint upregulation also provides rationale to overcome tumor immune resistance by the use of a PD-1 pathway inhibitor[103].

Figure 2. Combination epigenetic and immunotherapy in AML.

Proposed mechanisms of response to AML blasts in the setting of azacitidine (AZA). Increased expression of endogenous retroviral elements (ERV), IFN signaling, neoantigen presentation on MHC, and PD-L1 expression are shown. Antibodies against PD-L1 (e.g. atezolizumab, avelumab) with epigenetic therapy as a potential combination to target AML blasts and drive clonal expansion of tumor-specific T cells. (Adapted with permission from Stahl M, Curr Oncol Rep (2019) 21: 37)

Animal models have previously demonstrated a rationale for combination HMA and immune checkpoint inhibition. In a mouse model of metastatic breast cancer resistance to combined checkpoint blockade, co-treatment with azacitidine and the HDAC inhibitor entinostat with immune checkpoint inhibitors markedly improved treatment responses, curing more than 80% of tumor-bearing mice[96]. Additional studies demonstrated that this combination reduced numbers of myeloid-derived suppressor cells (MDSCs), suggesting that immune modulating effects of HMAs may potentially facilitate immunotherapy. Similar findings were observed in a mouse model of ovarian cancer, where the combination of decitabine and anti-CTLA-4 promoted differentiation of naive T cells into effector T cells, prolonging CTL responses as well as overall survival[97]. Clinical data in relapsed AML after transplant has demonstrated that the combination of azacitidine and donor lymphocyte infusion could induce long-term remissions, further supporting the concept of HMA-mediated facilitation of tumor-specific immunity[104].

Daver and colleagues systematically evaluated this question in an open label phase II study treating 70 relapsed/refractory AML patients with azacitidine and nivolumab[107]. Here, they demonstrated an overall response rate of 33% which is favorable to historical controls with other HMA combinations ranging between 10-15% in this setting. Of note, HMA-naïve patients, in addition to high levels of marrow infiltrating CD3 T cells at baseline had more favorable outcomes. Biomarker analysis also identified upregulation of CTLA-4⁺ CD4⁺ T cells (presumably Tregs) in non-responding patients. Although these combinations are promising in concept, the question remains to which patients would this provide the most benefit? We described the overt immunosuppressive state that can be established by increasing levels of disease (e.g. pDC depletion, T cell exhaustion, abortive tolerance), and would propose that patients with MRD would be the optimal population for treatment with immunotherapy.

The Emerging Importance of MRD in AML

The presence of minimal residual disease (MRD) after chemotherapy has emerged as a powerful prognostic factor in AML[12,108–112]. To this day, there are no approved therapies for AML patients with MRD. Due to differences in disease biology as well as patient age, the prognosis for older adults with MRD is particularly poor. Only 13% of AML patients over 60 with MRD after induction chemotherapy are alive and disease-free at five years[113]. In the setting of MRD, additional chemotherapy is often ineffective. It remains unclear if any post-remission therapy in older adults is beneficial[114], and even allogeneic stem cell transplantation (alloSCT) cannot rescue most adults with MRD after chemotherapy[12,108–111]. MRD positivity after consolidation chemotherapy in AML predicts early relapse and poorer outcomes with alloSCT, with one recent study indicating that transplant with MRD results in equivalently poor outcomes as transplant with active AML[12,108–111,113]. Data for monitoring AML MRD after transplant are limited, and there is currently no standard practice for post-transplant MRD monitoring. However, existing data indicate that the long-term prognosis for patients with AML MRD after alloSCT is also poor. AML MRD detection at day 100 post-transplant predicts relapse and poorer overall survival, and MRD detection at 1 year post-transplant may be an even more powerful negative prognostic marker[115–117]. In one recent study, patients who were MRD positive a year after alloSCT relapsed and died, while MRD-negative patients remained in CR[115].

Azacitidine with or without Checkpoint Blockade for Treatment of Minimal Residual Disease (MRD)

Single-agent Azacitidine has been used for the treatment of MRD in NPM1-mutant AML with molecular minimal residual disease[118]. Data are limited, but one case series describes 10 patients with normal karyotype AML and NPM1 mutations in first or second complete remission but with PCR detectable NPM1 consistent with MRD after intensive pretreatment with chemotherapy, allogeneic, or autologous HSCT. In this series, 7 out of 10 NPM1 MRD positive patients who underwent preemptive treatment with Azacitidine had a molecular response with an at least 1-log decrease in NPM1 MRD level. Five patients responded within 3 cycles of Azacitidine, and another 2 patients responded after cycle 4 and 5. However, some responses were only temporary. Three patients did achieve a sustained molecular complete response, but 2 out of these 3 patients underwent subsequent alloSCT. Intriguingly, 3 out of the 7 responders had previously undergone prior allogeneic SCT. Of the 3 patients with NPM1 MRD after prior allogeneic SCT, 1 achieved complete molecular response with clearance of MRD after 4 cycles, suggesting a potential enhancement of graft versus leukemia effect. Importantly, Azacitidine was well tolerated in this population, with reversible neutropenia and thrombocytopenia noted as the most frequent serious adverse events in 80% and 40% of patients, respectively.

Based on this and other data, Azacitidine has been used for treatment of MRD following alloSCT[119,120]. In the phase II RELapse prevention with AZAcitidine trials (RELAZA and RELAZA2), patients who experienced a decrease in CD34⁺ donor chimerism in the peripheral blood to values below 80% without concurrent hematologic relapse were treated with four cycles of Azacitidine. These studies demonstrated that pre-emptive therapy with azacitidine can prevent or substantially delay relapse in MRD+ patients. Azacitidine in this setting is well tolerated, with reversible grade ¾ neutropenia and thrombocytopenia and no significant exacerbations of GVHD. Lower-dose azacitidine has also been studied as maintenance therapy in AML patients during the first year after allogeneic transplant, with one study in 37 patients indicating that induction of a CD8⁺ T-cell response was associated with a reduced risk of relapse and improved relapse-free survival[121]. HMA treatment has also been previously shown to expand Tregs helping reduce GVHD while preserving this potential CD8⁺ T cell mediated GVL effect[122].

Several other post-transplant approaches with immune modulating agents have also been explored. Firstly, the use of the tyrosine kinase inhibitor, sorafenib, has been shown in pre-clinical models to increase levels of IL-15 and improve the phenotype (high levels of BCL-2 and decreased PD-1 levels) and function of leukemia-specific CD8 T cells. This increase in IL-15 levels is also mirrored in patients with FLT3-ITD⁺ AML responding to sorafenib post-alloHCT[123]. Additional studies with next-generation FLT3 inhibitors are ongoing. Secondly, for those patients without a targetable lesion such as FLT3 or IDH½, the use of the histone deacetylase (HDAC) inhibitor panobinostat has demonstrated impressive results in early phase studies [124]. While preclinical data support immunomodulatory activity and enhanced leukemia-specific activity with HDAC inhibitors, corollary studies are lacking from these early phase trials. Finally, the combination of HMA with the immune modulatory drug (IMiD) lenalidomide led to clinical responses in approximately half of patients treated at the recommended dose (25 mg daily with AZA 75 mg/m2 for 7 days)[125]. This did not, however, reverse T cell dysfunction in this population suggesting lenalidomide exerts its effects through other mechanisms.

In further support of a prominent role of immune-mediated control of AML both pre- and post-transplant, it was recently demonstrated that there is a significant downregulation of several MHC class II genes in patients who relapse following alloHCT[126]. This distinct mechanism of resistance has striking parallels to the recently described loss of MHC heterozygosity or beta-2-microglobulin (B2M) and immune escape in solid tumors during evolution or while receiving checkpoint blockade[127,128], supporting use of these novel agents in the treatment of AML.

Although checkpoint blockade could be a promising treatment approach for AML, we suggest that the patient population and specific immunotherapeutic agents will need to be optimized. First, lack of significant efficacy in alloHCT as disease burden increases suggest that PD-1/PD-L1 treatment in this same cohort may be ineffective secondary to the lack of DCs available for efficient cross-presentation in the setting of measurable disease. We propose that in order to overcome the abortive tolerance early on in the disease process secondary to engagement of CD8 T cells by expanding blasts, a more optimal approach may be to combine checkpoint blockade with HMA in the setting of MRD (Figure 2). This also provides a therapeutic option for many elderly patients unfit for further chemotherapy or allograft. This may also be an alternative treatment strategy for patients post-alloSCT who demonstrate evidence of disease reemergence, loss of donor chimerism, and need for additional therapies in combination with HMA or DLI.

Conclusion

We have much to learn regarding the use of immune modulating drugs for the treatment of patients with AML. Building on the expanding knowledge of immune suppression and dysfunction in AML patients remains a priority in the field. While we continue to rely on alloSCT as a cornerstone of treatment for patients with high-risk AML, it is reasonable to strive for alternatives that also engage a polyclonal immune response against chemotherapy-resistant cells. Cellular therapy will likely continue to play an important role, while the use of off-the-shelf NK cells or CAR-T cells becomes a more translatable approach. Although vaccine therapy remains an attractive concept in engaging both T- and B-cell mediated immunity following induction therapy, larger and more definitive trials are needed. Passive humoral immunity through the use of monoclonal antibodies will likely remain an active area of investigation in the field, given the ease at which these reagents can be generated and long track record of safety in hematologic malignancies. As novel cell-surface targets continue to be identified, these will need to be tested in early phase trials and hopefully moved to the upfront setting with combination chemotherapy. In the near future, larger trials testing immune checkpoint blockade alone or in combination with HMAs will be completed and we await these results. Evaluation of correlative studies from these patients remains critical as we attempt to better and understand and improve AML immunotherapy. Finally, the adverse prognostic significance of MRD remains an unmet need for therapy. We propose that AML MRD serves as an ideal target for combination therapies including HMA and checkpoint inhibition.